A GAIL-MAPPO Design for Coverage Path Planning of Spraying Robots

The paper addresses the coverage path planning problem of spraying robots. Based on the deep reinforcement learning and Generative Adversarial Imitation Learning(GAIL) theory, a novel GAIL-MAPPO algorithm is designed, where GAIL is used in early stage to fit expert experience, and Multi Agent Proximal Policy Optimization (MAPPO) then aims to further exploration. The issue of reliance on sequential actions of spraying is addressed by RNN. The proposed algorithm has demonstrated notable benefits in practical spraying operations, efficiently addressing constraints of the spraying and exhibiting adaptability to diverse spraying conditions. It is showing that the convergence time of GAIL-MAPPO is 38.5% of MAPPO, the average distance traveled is reduced by 34 units, and the path smoothness is improved by 18%, which is much higher than it of DFS, A*, and STC.